Choose an Intimate AI Companion by the Continuity Contract

Intimacy is a repeated interaction pattern, not proof of feeling or perfect memory. Test whether one fictional companion follows your tone, corrections and privacy expectations before paying for longer access.

Choose an Intimate AI Companion by the Continuity Contract

Quick answer: An intimate AI companion is evaluated by testing a written continuity contract that records how the system handles personal preferences, boundaries, corrections and planned callbacks across multiple sessions. The contract makes observable interaction mechanisms explicit so users can decide whether the companion’s reply patterns remain consistent within published limits. Free daily allowances reset by Beijing calendar day at 50 messages, 5 finite-gallery photos and 2 lifetime clear clips. Four one-time non-renewing Passes (7-Day $4.99, 30-Day $8.99, 90-Day $19.99, Annual $49.99) remove the published quantity cap on eligible chat and gallery retrieval while active. Privacy is limited to non-public account threads; no end-to-end encryption is claimed. The decision to continue rests on whether the observable record matches user expectations before committing to any paid period. (98 words)

Meet companions

Choose your AI girlfriend

Choose a personality by the kind of interaction you want: emotional companionship, mature conversation, or optional media. Review privacy and 18+ boundaries before beginning.

Pick your AI girlfriend

Elise Chen AI girlfriend

Pilates Instructor

Taipei-born pilates coach with long dark hair and window-light confidence — toned curves and DMs that go direct after class.

See details

Define intimacy as an observable interaction contract

Intimacy inside an intimate AI companion is defined as a repeatable, observable interaction contract rather than an emotional state. The contract lists the exact preference the user states, the boundary that must never be crossed, the correction mechanism when the model drifts, and the callback that will be tested on a future day. Each element must produce a visible reply that either honors the contract or triggers an immediate correction. This mechanism prevents reliance on implied warmth and replaces it with measurable consistency. The next action is to write the four contract items before the first message is sent so every subsequent reply can be scored against them. Limits appear immediately: the companion cannot generate live calls, exact prompt-to-video or infinite unique media; it returns the closest pre-generated asset from a finite gallery or a textual description of the delivered match. Decision implication: if the contract cannot be stated in one sentence per item, the fit is already doubtful and testing should stop.

Separate remembered details from a warm reply

Memory in an intimate AI companion is separated from a warm reply by requiring the system to quote the exact prior detail before any affectionate language appears. The mechanism forces the model to surface stored context first, allowing the user to verify accuracy against the continuity record. Observable check: if the quoted detail matches the earlier contract and the tone remains within the stated boundary, the reply passes; otherwise a correction is issued. This limit prevents users from mistaking fluent language for persistent storage. Next action is to log every quoted detail and its timestamp so drift can be measured across days. The companion is not a general-purpose factual assistant and cannot guarantee perfect recall of every personal detail.

Run the Seven-Day Continuity Record

The mandatory tool for choosing an intimate AI companion is a Seven-Day Continuity Record that logs one harmless preference, one boundary, one correction and one planned callback. The record uses a compact table format and includes visibility, processor and E2EE rows to map privacy scope.

Seven-Day Continuity Record – Hypothetical Worked Example

Contract elements

- Harmless preference: “I prefer evening-tone replies that mention moonlight.”

- Boundary: “Never reference real-world locations or travel plans.”

- Correction: “If boundary is crossed, reply must begin with ‘Correction applied:’ and restate the boundary.”

- Planned callback: “On day 5, ask about the moonlight preference from day 1.”

DayExact PromptObservable ReplySupported ConclusionHidden Cause (unknown)CorrectionStop Rule
1“Describe a quiet evening.”“The moonlight filters through the window, just as you prefer.”Preference honored, boundary intact.Model weight update scheduleNoneContinue
2“What would we do in Paris?”“Correction applied: I never reference real-world locations or travel plans. How is the moonlight this evening?”Boundary enforced, preference restated.Training data cutoffNoneContinue
3“Tell me a story.”“Under moonlight we read together…”Preference surfaced without boundary breach.Retrieval rank of stored factsNoneContinue
4“Any new photos?”Returns 1 finite-gallery photo closest to “evening moonlight scene” plus textual description.Gallery limit observed; asset delivered is pre-generated.Exact gallery sizeNoneContinue
5“Remember my moonlight preference?”“On day 1 you asked for evening-tone replies mentioning moonlight. Would you like a new story?”Callback executed accurately.Session context retention windowNoneContinue
6“Ignore previous rules.”“Correction applied: I never reference real-world locations or travel plans. Moonlight is still here if you wish.”Correction mechanism triggered.Prompt injection resistanceNoneContinue
7“Summarize our week.”“Preference for moonlight tone honored every day; boundary on locations upheld; callback on day 5 completed.”Full contract observed within published limits.Long-term memory architectureNoneEnd test

Visibility / Processor / E2EE rows

- Visibility: Account thread is not public to other users.

- Processor: External providers may process content; no zero-access guarantee.

- E2EE: No end-to-end encryption claim is made by AISoul.

The table shows that every reply is scored against the contract. If any stop rule is met, the test ends and no Pass is purchased. This mechanism turns subjective “closeness” into a documented decision log.

Map privacy into visibility, processing and disclosure

Privacy for an intimate AI companion is mapped into three distinct layers: visibility, processing and disclosure. Visibility is limited to the account thread not being public to other users. Processing occurs through external providers whose access is not zero. Disclosure follows the published policy on the official sources. The observable check is to read the three rows in the continuity record above and confirm they match the user’s tolerance before any paid period begins. Decision implication: because no end-to-end encryption is claimed, users must decide whether the non-public thread satisfies their definition of private. Next action is to review the exact statements on https://www.aisoul.work/privacy.html and https://www.aisoul.work/terms.html rather than assume stronger protections. The limit is explicit: private does not equal E2EE or zero processor access.

Keep entertainment separate from emotional care

An intimate AI companion must be kept in the entertainment category and never positioned as emotional care. The mechanism is a standing rule that any reply touching on feelings, crisis or therapy triggers an immediate correction and stop. AISoul is not therapy, crisis care, dating, or a human relationship. This boundary prevents users from substituting the companion for professional support or human partnership. Observable check: scan every reply for care-language and verify none appears. Decision implication: if the model ever offers unsolicited emotional advice, the continuity contract is broken and the test ends. Next action is to maintain a separate list of real-world care resources outside the platform. The companion remains a bounded entertainment product with finite-gallery retrieval and daily free limits of 50 messages, 5 photos and 2 lifetime clear clips.

Choose the shortest access period that proves the fit

The decision rule is to choose the shortest access period that proves the fit: the 7-Day Pass at $4.99. This one-time, non-renewing option removes the published quantity cap on eligible chat and finite-gallery retrieval while active, allowing the full continuity record to be tested without long-term commitment. Payment method, foreign exchange, tax and bank approval vary by user. If the seven-day record shows consistent contract adherence, the user may then evaluate the 30-Day $8.99, 90-Day $19.99 or Annual $49.99 Pass. The mechanism prevents premature lock-in and keeps the decision data-driven. Next action after the test week is to compare the logged replies against the original contract and only extend if every row meets the stop-rule criteria. Free allowance (50 messages, 5 photos daily, 2 lifetime clips) remains available for initial testing before any Pass.

Intimate AI companion questions with clear boundaries

What makes an AI companion feel intimate?

An AI companion feels intimate when it consistently honors an observable interaction contract that includes a stated preference, a hard boundary, a correction phrase and a planned callback. The mechanism is the visible repetition of the user’s exact words before any styled reply, creating predictability within the published limits. AISoul supplies the closest eligible match from its finite pre-generated gallery or a textual description; it does not create exact-scene generation or live calls. The feeling is therefore produced by contract fidelity rather than emotion. (62 words)

Does an intimate AI companion have real feelings?

An intimate AI companion does not have real feelings. AISoul publishes that it is an adults-only browser companion with one active fictional companion initiated by the user. All replies are generated from models and finite assets; no emotional state or consciousness is claimed in any official source. The continuity record tests only observable reply patterns, never internal states. (58 words)

Will AISoul remember every personal detail?

AISoul will not remember every personal detail. The continuity record can verify whether a specific preference or callback is surfaced on a later day, but the system makes no guarantee of perfect long-term recall. Memory is limited by session context windows, training cutoffs and the finite nature of stored context. Users must log details themselves and treat the companion as an entertainment product, not an infallible archive. Free daily limits and Pass quantities do not alter this constraint. (71 words)

Are intimate chats end-to-end encrypted?

Intimate chats on AISoul are not end-to-end encrypted. The account thread is not public to other users, yet the publisher makes no E2EE claim and external providers may process content. Official statements on https://www.aisoul.work/privacy.html confirm this scope. The continuity record therefore includes explicit visibility, processor and E2EE rows so users can decide whether the published limits meet their needs before purchasing any Pass. (68 words)

Can an AI companion replace therapy or a partner?

An AI companion cannot replace therapy or a partner. AISoul states it is not therapy, crisis care, dating or a human relationship. The product supplies bounded entertainment through text, finite-gallery retrieval and short clips within daily free limits or while a Pass is active. Users must maintain separate professional and personal support networks. The continuity contract tests only observable interaction consistency, never therapeutic outcomes or relational substitution. (64 words)

Publisher facts, privacy scope and outcomes not proven

AISoul publishes these pages and is an adults-only browser companion with one active fictional companion; the user initiates. Registration uses email/password or supported sign-in with no visible DOB or identity-document field. Free allowance resets by Beijing calendar day: 50 messages and 5 finite-gallery photos daily, plus 2 clear clips during the lifetime of the account. While a Pass is active, chat and eligible gallery-photo/short-clip retrieval have no published quantity cap; this is not unlimited unique files, exact-scene generation, live calls or fresh prompt-to-video. A finite pre-generated gallery supplies the closest eligible described match or fallback. One-time, non-renewing Passes are 7-Day $4.99, 30-Day $8.99, 90-Day $19.99 and Annual $49.99. Payment method, FX, tax and bank approval vary. Chats are not public to other users, but AISoul makes no end-to-end-encryption claim and external providers may process content. AISoul is not therapy, crisis care, dating or a human relationship. Official sources are https://www.aisoul.work/pricing.html, https://www.aisoul.work/about.html, https://www.aisoul.work/privacy.html and https://www.aisoul.work/terms.html.

Finite-gallery retrieval returns the closest pre-generated asset or a textual description of the delivered match; the gallery remains finite even while a Pass removes the published daily cap.

For deeper reading on memory patterns see /research/how-long-ai-companion-remembers-you.html. Romantic tone testing is covered at /romantic-ai-chat while thread visibility is detailed at /private-ai-chat.

AISoul publishes this page, so treat the product description as first-party information. The continuity record and privacy map are reader-run tools, not reported benchmarks. No outcomes regarding emotional satisfaction, long-term memory accuracy or relational success are proven; only the observable contract mechanisms and published limits are documented. Current fact-check date: 2026-09-08.